Subscribe to hangytong
Subscribe to hangytong
Share Dialog
Share Dialog
<100 subscribers
<100 subscribers
Hardware
GPU -> has reached a stage whereby the cost of compute is significantly low enough to provide improved margins to the end users.
Chips -> transistors are becoming smaller, improving performance.
Machinery -> Precision manufacturing is constantly improving with new tech to enable smaller transistors with good yields.
Data storage
AI cannot exist without data centers. Demand will continue to rise.
Infrastructure is key. Needs power grids, cooling source, etc.
Cloud Computing
This is where all your models, training and inference comes in. its the software built on top of the data centers to provide current AI capabilities.
Solves:
Operational Efficiency - automation
Enhanced Decision Making - able to draw insights faster and maybe better from a wider cast of datasets
Reduced cost
AI is scalable and fast
Expected margin improvements:
15 - 30% on operational costs
20 - 40% on labor costs
Current improvement in tech to enable AI seems to be flowing moore's law.
Infrastructure to enable such technology (data centers) have also seen a significant increase in spending.
Government and private companies are doing all they can to race against building the best infra out there for AI, so the rate at which new and interesting AI products and applications churn out should increase as well. Since infrastructure is mostly a solved game now, with enough data, AI applications will start to see major shifts.
As a whole, we can think of the entire stack in AI summarized into 3 parts:
Data -> Infra -> Application
Web2 applications heavily relies on the central providers (AWS, GOOGLE, AZURE) to provide compute. To build your own AI application from scratch, not only would you need to house your own data, you would need to get your own hardware GPU, though it can be rent from other data centers, and you would also need to build the entire infrastructure yourself. Not only that, you somehow also need to get a good enough dataset to build a good model. Most would not be able to do that, unless you have years of data waiting for you. In web2, its a winner takes all situation. Those with the best datasets and infrastructure wins.
This summarizes into the following problems for smaller players:
Infrastructure support : GPU & Storage
Data
In order for web3 to work, it needs to help to solve those 2 problems above while also helping to provide equivalent if not better metrics than web2 in terms of application, in which we can also summaries to the following:
scalability
cost effectiveness
Speed
In web3, infrastructure is mostly decentralized in nature. While it helps to solve the issue of improving cost effectiveness while remaining scalable, it presents a problem whereby some tech stacks that requires nodes to do the inference may need some sort of verifiability to know that they output is as intended and not rugged into something else desirable.
A new vertical in crypto that can help to solve certain centralization issues is privacy. TBH, most companies would never need it. However, for certain applications which are data sensitive, this provides that option (FHE, TEEs etc).
** note that verifiability and privacy are 2 important expect of crypto but yet it cannot compromise the 3 important factors (scalability, cost effectiveness, speed) else it would just be over engineering with a "good to have" use case.
Important to note these 2 things:
AI in crypto. Where AI helps to solve certain things in crypto
Crypto in AI. Where crypto helps to solve/improve things in AI.
At first sight, AI in crypto might not seem bullish, but i think this is where the hype is at. This 100% does not help to solve the application problems listed above, but it has the potential to create MEME and value to crypto itself. Agents are the best example, technically the entire infrastructure could be ran on a centralized system running python scripts to create the agents, put a token on top of it and boom, hype is here. Agents are now also helping in defi to automate optimizing yields. Of course, the infra can be built using decentralized infrastructure in crypto that amalgamate resources (kuzco, heurist, etc) to build the agent and yet also give a very good and consistent output, then that's truly bullish on crypto.
Crypto in AI is where certain things in AI itself cannot happen without crypto. The simplest example of this is privacy where cryptography itself is the key. However, although this is a bullish vertical in itself, like any other technology, it will become commoditized and competition will flood in. Its imperative to only find competent founders and companies who could be in the "winner takes all" list. Verifiability is also another aspect which allows AI systems to demonstrate their integrity and trustworthiness, fostering confidence in their decision-making capabilities. There also lies an opportunity in supply chain management where transparency and security is much needed in logistics and inventory management. Coupled with AI, it could bring about effective use of data and actionable results.
Hardware
GPU -> has reached a stage whereby the cost of compute is significantly low enough to provide improved margins to the end users.
Chips -> transistors are becoming smaller, improving performance.
Machinery -> Precision manufacturing is constantly improving with new tech to enable smaller transistors with good yields.
Data storage
AI cannot exist without data centers. Demand will continue to rise.
Infrastructure is key. Needs power grids, cooling source, etc.
Cloud Computing
This is where all your models, training and inference comes in. its the software built on top of the data centers to provide current AI capabilities.
Solves:
Operational Efficiency - automation
Enhanced Decision Making - able to draw insights faster and maybe better from a wider cast of datasets
Reduced cost
AI is scalable and fast
Expected margin improvements:
15 - 30% on operational costs
20 - 40% on labor costs
Current improvement in tech to enable AI seems to be flowing moore's law.
Infrastructure to enable such technology (data centers) have also seen a significant increase in spending.
Government and private companies are doing all they can to race against building the best infra out there for AI, so the rate at which new and interesting AI products and applications churn out should increase as well. Since infrastructure is mostly a solved game now, with enough data, AI applications will start to see major shifts.
As a whole, we can think of the entire stack in AI summarized into 3 parts:
Data -> Infra -> Application
Web2 applications heavily relies on the central providers (AWS, GOOGLE, AZURE) to provide compute. To build your own AI application from scratch, not only would you need to house your own data, you would need to get your own hardware GPU, though it can be rent from other data centers, and you would also need to build the entire infrastructure yourself. Not only that, you somehow also need to get a good enough dataset to build a good model. Most would not be able to do that, unless you have years of data waiting for you. In web2, its a winner takes all situation. Those with the best datasets and infrastructure wins.
This summarizes into the following problems for smaller players:
Infrastructure support : GPU & Storage
Data
In order for web3 to work, it needs to help to solve those 2 problems above while also helping to provide equivalent if not better metrics than web2 in terms of application, in which we can also summaries to the following:
scalability
cost effectiveness
Speed
In web3, infrastructure is mostly decentralized in nature. While it helps to solve the issue of improving cost effectiveness while remaining scalable, it presents a problem whereby some tech stacks that requires nodes to do the inference may need some sort of verifiability to know that they output is as intended and not rugged into something else desirable.
A new vertical in crypto that can help to solve certain centralization issues is privacy. TBH, most companies would never need it. However, for certain applications which are data sensitive, this provides that option (FHE, TEEs etc).
** note that verifiability and privacy are 2 important expect of crypto but yet it cannot compromise the 3 important factors (scalability, cost effectiveness, speed) else it would just be over engineering with a "good to have" use case.
Important to note these 2 things:
AI in crypto. Where AI helps to solve certain things in crypto
Crypto in AI. Where crypto helps to solve/improve things in AI.
At first sight, AI in crypto might not seem bullish, but i think this is where the hype is at. This 100% does not help to solve the application problems listed above, but it has the potential to create MEME and value to crypto itself. Agents are the best example, technically the entire infrastructure could be ran on a centralized system running python scripts to create the agents, put a token on top of it and boom, hype is here. Agents are now also helping in defi to automate optimizing yields. Of course, the infra can be built using decentralized infrastructure in crypto that amalgamate resources (kuzco, heurist, etc) to build the agent and yet also give a very good and consistent output, then that's truly bullish on crypto.
Crypto in AI is where certain things in AI itself cannot happen without crypto. The simplest example of this is privacy where cryptography itself is the key. However, although this is a bullish vertical in itself, like any other technology, it will become commoditized and competition will flood in. Its imperative to only find competent founders and companies who could be in the "winner takes all" list. Verifiability is also another aspect which allows AI systems to demonstrate their integrity and trustworthiness, fostering confidence in their decision-making capabilities. There also lies an opportunity in supply chain management where transparency and security is much needed in logistics and inventory management. Coupled with AI, it could bring about effective use of data and actionable results.
hangytong
hangytong
No activity yet